How did you run the 100000 iterations, it seems it will take a week to train for only 1 setting about trident HOT 1 CLOSED

Hey Michalis,
Thanks for your kind words, appreciate it!
Indeed, we have set the maximum epochs to 100,000 in all training configs but we also mention in the paper and the repository Readme.md that "We obtained the best model at the 82,000-th and 67,500-th iteration for (5-way, 1-shot) mini and tieredImagenet tasks respectively, and at the 22,500-th and 48,000-th iteration for (5-way, 5-shot) mini and tieredImagenet tasks, respectively." Thus, it actually doesn't take 6-7 days for training to finish. Regarding the parallelization of code, due to the scarcity of meta-gradient / meta-update computation libraries in pytorch, we chose to use the 'learn2learn' implementation which doesn't offer parallelization functionalities.
Thanks for your interest in our work and for the apt question, I think it makes sense to change this epoch parameter in the config files.

from trident.